Control Algorithms for a Sailboat Robot with a Sea Experiment

Transcription

1 Control Algorithms for a Sailboat Robot with a Sea Experiment Benoit Clement To cite this version: Benoit Clement. Control Algorithms for a Sailboat Robot with a Sea Experiment. Conference on Control Applications in Marine Systems, Sep 213, Osaka, Japan <hal-87579> HAL Id: hal Submitted on 7 Oct 213 HAL is a multi-disciplinary open access archive for the deposit and dissemination of scientific research documents, whether they are published or not. The documents may come from teaching and research institutions in France or abroad, or from public or private research centers. L archive ouverte pluridisciplinaire HAL, est destinée au dépôt et à la diffusion de documents scientifiques de niveau recherche, publiés ou non, émanant des établissements d enseignement et de recherche français ou étrangers, des laboratoires publics ou privés.

2 Control Algorithms for a Sailboat Robot with a Sea Experiment Benoit Clement LabSTICC, UMR CNRS 6285, ENSTA Bretagne, 2 rue Francois Verny, 2986 Brest Cedex 9, France. Abstract: A sailboat robot is a highly nonlinear system but which control is relatively easy, however. Indeed, its mechanical design is the result of an evolution over thousands of years. This paper focuses on a control strategy which remains simple, with few parameters to adjust and meaningful with respect to the intuition. A test on the sailboat robot called Vaimos is presented to illustrate the performance of the regulator with a sea experiment. Moreover, the HardWare In the Loop (HWIL) methodology has been used for the validation of the embedded system. Last point is that this HWIL simulation compared to the real experiment is also a confirmation that the dynamic model used for control is correct. Keywords: robotics; autonomous vehicles; nonlinear control; sailboat; marine systems; experiment 1. INTRODUCTION The development of autonomous surface vehicle (ASV) is motivated by several civil and military applications: oceanographic research, marine charts development, meteorological data collecting for civil part, and operations of coastal protection, port monitoring and demining for the military part. These devices are generally motorized ASVs and they have been studied for a long time but have strong energy limitations. The sailboats have a number of advantages over boats powered by motors: energy selfsufficient in the first point but also the fact of being noiseless (which can be an asset for military applications or biodiversity observation). Romero-Ramirez [212] and Stelzer [212] PhD Thesis propose a complet state of art about ASV with historic perspectives of the main issues. 1.1 Sailboat robots While sailing has a long tradition, robotic sailing is a fairly new area of research. Indeed, first references concerning automatisation for sailing is in Warden [1991] and the projects initiated in the late 9 show that it is a young issue (Aatrijk et al. [23], Elkaim [29], Neal [26]). On one hand, the sail propulsion allows long range and long term autonomy; on the other hand, the dependency on changing winds presents a serious challenge for short and long term planning, collision avoidance, and boat control. Moreover, building a robust and seaworthy sailing robot is not a simple task, leading to a truly interdisciplinary engineering problem. The characteristics of a sailing boat can be defined as follows (Stelzer and Jafarmadar [211]): wind is the only source of propulsion it is not remotely controlled; the entire control system is on board it is completely energy self-sufficient Sailing robot are efficient solutions for fuel saving on any boat; Schlaefer and Blaurock [211] have summarized the state of art on this subject. Various aspects of system design and validation are discussed, further highlighting the interdisciplinary nature of the field. Methods for collision avoidance, localization and route planning are covered but few papers are dedicated to the automatic control aspects. Indeed, most of rudder control laws are based on Propotional Integral and Derivative (PID) heading control with overshoot, oscillations problem and difficulty to reach a waypoint. This is the main reason which leads to a reflexion about the change of strategy, the line following defined later in the paper. It is important to notice that the development of autonomous sailboat is mainly due to robotic challenges such as Sailbot 1, Microtransat 2 or World Robotic Sailing Championships 3 and the accompanying International Robotic Sailing Conference (IRSC) which provides a venue to discuss the broad range of scientific problems involved in the design and development of autonomous sailboats (Collective [29], Schlaefer and Blaurock [211]). 1.2 Vaimos robot As mentioned before, the ASV development is a multidisciplinary problem; this paper only focus on the algorithmic aspects with the comparison between an HWIL (Hardware In The Loop) simulation and a real experiment. This has been performed with the autonomous sailboat robot VAIMOS developed by IFREMER 4 and ENSTA Bretagne 5 for oceanographic issues as presented in Menage et al. [211]. Indeed, as mentioned in Thomas et al. [211] recent publications revealed that the mixed 1 website: 2 website: 3 website: 4 French Research Institute for Exploration of the Sea 5 Ecole Nationale Superieure des Techniques Avancees de Bretagne

3 1.4 Organization of the paper The aim of this paper is to validate the control algorithms at two levels: simulation allows to show if the behaviour is as expected, and experiment allows to definitly validate that the algorithms are robust. The experiment compared to the simulation gives some indications about the model validity. Considering the VAIMOS sailboat, this paper is organized in three parts. First a simplified model for simulation and the real sailing robot are presented; then the algorithms used for the robot guidance, navigation and control (GNC) are developed and discussed. The last part is dedicated to comparison between the HWIL simulation and the experiment at sea. 2. THE BOAT : SIMULATOR AND HARDWARE Fig. 1. Vaimos sailboat and another sailboat robot layer may present surface singularities for bio-geochemical parameters (temperature, salinity, turbidity, chlorophyll). Those studies question the common view of a homogeneous mixed layer. However, the degree of ubiquity of these surface singularities and their horizontal structures remains largely unknown because of the lack of adequate instrumentstosamplethefirstcentimetersoftheocean.inorder to be able to document the gradients of several parameters between the top centimeters and the sub-surface of the ocean, wan autonomous sailboat has been developed which is able to sample the ocean surface at two depths (the first 1 centimeters and about 1 meter). The sailboat is shown on figure GNC (Guidance, Nagigation, Control), Simulation, Experimentation As in any robotics activities, the simulation takes an important part for hardware and software validation but it is even more important when the experiment is not so easy to perform as for a marine robot. In these conditions, choosing an appropriate model is essential for success. As mentioned in Fossen [1995], and the references inside, the precise modeling is an difficult task due to fluidsolid interactions and do not answer to the robotician need which is a macroscopic model. In the literature, one can found various analytic models for sailboats control as Yeh and Binx [1992], Elkaim [29], Briere [28], Cruz and Alves [21] but a simple state space representation has been chosen; it is based on Lagrangian Mechanics proposed in Jaulin [24] with smooth modifications. The model equations are described in a section dedicated to modeling. This model has been used to develop the control algorithms but also to perform the HWIL simulation to prepare the sea test. Moreover, the real robot is described, particularly the sensors and actuators with a comparison between the model and the robot. Modeling a boat in details, and even more a sailboat, is a real challenge. As mentioned in Astrom and Murray [211], a model is a representation of a system dynamics used to answer questions via analysis and simulation. The chosen model depends on the questions we wish to answer, and so there may be multiple models for a single dynamical system as mentioned in the state of art for sailboat robots, with different levels of fidelity depending on the phenomena of interest. This section presents a model that can be used for a macroscopic behavior of a sailboat. The experiment will show that this model is efficient for performing HWIL simulations. This section gives also a description of the real sailboat robot Vaimos. There exists lots of sailboat robots and the following references gives some examples Sauze and Neal [26], Rynne and von Ellenrieder [29], Collective [29], Schlaefer and Blaurock [211], Romero-Ramirez [212] and Stelzer [212]. This two last work (one is in French) are the more complete and up to date state of art on the subject. 2.1 Sailboat Modeling In this article, as the control issue is considered, the candidate model for sailboat is the one proposed in Jaulin [24], with the difference that the input commands are not the angle between the boat and the sail but the length of the sail sheet. This model has been implemented in a simulator under QT creator 6 for the control tuning and then for the experience simulation. There exist more complex simulators as mentioned in Romero-Ramirez [212] or proposed by Guillou [21] based on multiagent modeling but they are not adapted to control problem with state space point of view. The main advantage for our model is the simplicity: for interpretation, for running the simulation(alongmissiononlytakesfewseconds)andforcontrol tuning. The proposed model is based on the Newton s laws of motion applied to the sailboat presented in figure 2. As in space applications, the maritime environment is highly perturbated, and a very accurate model is useless for GNC. The comparision of the results of this model and the real experiement will cinfirm this hypothesis. The following state space equations are derived from dynamic and kinematic considerations: 6 website :

4 Name Description (x, y, θ) boat position and orientation v boat velocity ω rate of rotation f s wind force on the sail f r water force on the rudder δ v angle between the sail and the longitudinal axis a wind speed ψ wind orientation σ sheet tension indicator: if σ, there is no sheet tension, else the sail is efficient δ s maximum angle between the sail and the boat axis (i.e. the length of the mainsheet) δ r angle between the rudder and the boat axis Table 1. Model variables Name α d =.5 α v = 1 Description drift coefficient friction coefficient for longitudinal movement friction coefficient for rotational movement lift coefficient for the sail lift coefficient for the rudder α ω = 6 α s = 175 α r = 25 p 6 = 2m geometric property (see figure 2) p 7 = 2m geometric property (see figure 2) p 8 = 4m geometric property (see figure 2) M = 2kg mass J = 1kg.m 2 inertia Table 2. Model Parameters ẋ=vcosθ +α d acosψ ẏ=vsinθ +α d asinψ θ=ω v= f ssinδ v f r sinu 1 α v v 2 M ω= f s(p 6 p 7 cosδ v ) p 8 f r cosδ r α ω ω J f s =α s asin(θ ψ +δ v ) f r =α r vsin(δ r ) σ=cos(θ ψ)+cos(δ s ) { if σ π θ +ψ δ v = else sign(sin(θ ψ))δ s where the variables are presented in Table 1 and the numerical value of each parameter (coefficients, geometric properties) evaluated by experiments are in Table VAIMOS Robot The sailboat robot Vaimos objective is to collect biogeochemical parameters (temperature, salinity, turbidity, chlorophyll) as mentioned in Thomas et al. [211] and Menage et al. [211]; it is presented in figure 1. She has been built with a Miniji 7 hull adapted to be autonomous. The most important change is that the rig has been turned into a spirit rig; it has been done to control one actuators instead of two. The Miniji hull characteristics are given in Table 3. 7 Naval architect website: Fig. 2. Sailboat geometry Deplacement (kg) 16 Length (m) 3,65 Beam (m),86 Draft (m),65 Table 3. Miniji Specifications Fig. 3. Embedded system architecture Figure 3 shows the global architecture for the embedded system with the processor, the sensors and the actuators. The GNC system is composed by: 2 sensors: the main sensor is a weather station which provides: wind speed and direction, air temperature, atmospheric pressure, GPS position, heading; an IMU for reliable kinematic parameters; 2 actuators to control the sail and the rudder: a brushless motor for the rudder; a stepping motor to control the angulation of the sail; a processor with a serial data acquisition interface; power supply : two 12V marine batteries to deliver 12Ah with 24V voltage. To ensure safety during the experiments, it is always possible to have a remote control via a wifi connection.

5 End of navigation upwind navigation upwind portside navigation upwind starboard non feasible course non feasible course line validation Next line Fig. 5. Control loop Line following No more line Fig. 4. Petri net for navigation 3. GUIDANCE, NAVIGATION AND CONTROL As in the aeronautic and space applications (Imbert and Clement [24], Breivik and Fossen [27]), robustness is a key issue to ensure the robot mission. the robust control strategy is then performed in a classical way: Control to ensure the trajectory following. A line following is proposed instead of a heading following as it is often proposed. This is done by the rudder and sail tuning. Guidance for the trajectory definition : direct to a waypoint or by beating to windward according to wind direction Navigation is defined off line by a way points collection. One can imagine that obstacle avoidance has to be included in this part of GNC. Then for each part of guidance and control, some innovations are proposed: intuitive control laws, line following, pathway definition. 3.1 Navigation: motion planning During a mission, the robot has to follow successive segments denoted [a j,b j ],j {1,...,j max } such that a j+1 = b j if j < j max. Note that for a closed path, a 1 = b jmax. It is necessary to determine an angle ξ which is the limit angle that the sailboat can have with the wind. Then a course θ is considered as feasible if cos ( ψ θ ) cosξ > (1) according to the Figure 6. As proposed by Guillou [21], the mission is managed by a state machine; this Petri net is presented in Figure 4. The mission starts with the transition t 1 from p place (standby) to p 1 and the robot follow the first line (j = 1). Then according to the relative wind, it can be direct or with tacks. Then if cos(ψ γ ϕ j ) + cosδ prs < with ϕ j the angle defined by the segment [a j,b j ], it is necessary to adapt the navigation strategy with tacks as the sailboat is not able to follow the line. When beating to windward, the sail tuning is such that u 2 = δ smax = Fig. 6. Line following which means that the sail is closed at the maximum. A line is validated as soon as the waypoint b j is behind the boat: b j a j,m b j > and next line can be followed. 3.2 Sail and rudder control Theoretical methods have been published for sailboat regulation as Herrero et al. [21], but it is necessary to have a precise model. As mentioned in section 2, it is not realistic. Moreover, in order to adapt the algorithms to any robot, the navigation and the control have to be easy to implement and easy to tune with a physical meaning for tuning parameter. The proposed control architecture is given by the figure 5 answers this issue. The objectives for this control loop are: the boat speed optimization according to the wind, the line following according to the guidance algorithm demand. Line following. The most common approach is to consider a navigation by waypoint validation as mentioned in Romero-Ramirez [212] and references inside. To ensure a correct mission for measurements, the guidance is performed by a line following strategy. Indeed, due to drift and current, a heading strategy will fail because a sailboat is not able to navigate wind ahead. Rudder control. First consideration is that if cos γ <, the boat navigates to the opposite direction, then: δ r = δ rmax.sign(sinγ) (2) where δ rmax is the maximal rudder deflection. If the boat navigates in the good direction, cosγ >, the heading error is defined by e cap = sinγ. A proportional heading

6 Fig. 7. Mainsheet tuning (sail angle) control is defined by δ r = δ rmax.e cap. The line following strategy is a compromise between the distance to the line and the line heading. The distance to the line is given by: e = m a sinα, (3) where m is the center of gravity of the boat. Note that this allows to know which side of the line is considered. The following control strategy is used to follow the line: δ r = δ rmax.(λsinγ +(1 λ) sign(e)). (4) where λ is the tuning parameter for the compromise. For our example λ =.7 has been chosen in order to avoid oscillations around the line. Sail control The sail control tuning is based on the human behavior observation. Let us consider that β is the optimal angle when sailing with wind abeam (perpendicular to the boat). This angle is determined according to the type of boat, the sail and so on: this is the human part of the tuning. The maximum angle between the sail and the boat, denoted δ s, is a (ψ θ) function that can be modeled by a cardioid given by: δ smax = π ( ) q cos(ψ θ)+1 2. (5) 2 where q is positive and chosen in order to maximize the sail efficiency. This model is such that: whenψ = θ+π,theboatiswindaheadandthemodel gives δ smax = ; when ψ = θ, the boat is wind aft and the model gives δ smax = π 2 (the sail is open); when ψ = θ ± π 2, the boat is wind abeam and by definition, δ smax = β. This last item gives the value for q. Indeed equation 5 becomes: ( ) q π 1 2. = β (6) 2 which means q = log( π 2β ) log(2). (7) The δ vmax function is illustrated by figure 7. This kind of tuning has been validated by various simulations and tests as presented in Jaulin et al. [212] or the one presented in section 4.2. The main point of this approach is that it remains very simple and can easily be adapted to any sailboat. This demonstrator is used used for various experiments and then it is shown in this paper that the control aspects are useful to control the boat and to save power to increase autonomy. Fig. 8. Trajectories: objective in red and realized in green Fig. 9. HWIL architecture 4. SIMULATION AND EXPERIMENT These GNC algorithms are validated by HWIL simulations where the real embedded system is used and boat behavior and sensors data are simulated; it is then compared with the experimentation performed between Brest and Douarnenez in January, 17th, 212. The experiment to perform is a return trip from Brest to Douarnenez in Brittany as mentioned in Figure 4. This figure is a screen capture from the webpage generated by J. Nicolas for the test: Simulation HardWare In the Loop (HWIL) simulation is performed according to the Figure 9. For these simulations, the sailboat model is described in section 2.1 and the embedded system is the real one. The sea state is not emulated and the wind is considered as constant with white noise. This noise can be considered as including the sea state effect and the current. The results are presented in figure 1. The wind data are given in direction and intensity; the trajectory and the reference are superposed (with a zoom)

7 wind dir ( ) wind speed (m/s) Waypoints 5 and trajectory1 15 Zoom 2 on WP wind dir ( ) wind speed (m/s) Waypoints 2and trajectory Zoom on 1 WP latitude ( ) latitude ( ) attitude ( ) longitude ( ) roll pitch yaw attitude ( ) longitude ( ) roll pitch yaw Rudder deflection Sail opening Rudder deflection Sail opening.5 sin(u) sin(u).5.5 e (m) time (min) Fig. 1. Data acquisition from simulation and the waypoints are shown with crosses; the sailboat attitudes are also plotted with the control command (rudder deflection and sail opening). 4.2 Experiment Experiment and HWIL simulation have exactly the same waypoints set and the same embedded system. Note that the mission is not complete due to various troubles: the supervising boat had an engine problem that could not be fixed due to the night; the sailboat was not following the planned mission any more: a part of the mechanical control system was broken. In spite of this, the duration of the mission and the good behavior of the boat lead to very positive conclusions. The wind comes from the south and the robot was always at a distance less than 3 meters to its line except twice: to avoid collision with a military submarine coming back to Brest; to avoid collision with a fisherman boat. All details and data are available from the authors under request; they are not provided in this paper due to the big data. The results are presented in figure 11. The figure architecture is the same as Figure 1 to allows the comparision. During the mission and after, a dashboard is used to analyze all the log files produced by the embedded program. For example, near the end of the experiment, it has been seen that the sail angle measured by the weather station e (m) time (min) Fig. 11. Data acquisition from experiment is incoherent with the one deduced from the input. This was probably due to the mechanical problem in the sail control system that was discovered at the end. Because it was during the night, it was difficult to see the sail angle without this dashboard. 4.3 Discussion The comparison between Figure 1 and Figure 11 rises various comments: Even if the wind input is not similar, the real boat behavior is close to the simulated boat. Indeed, this due to wind filtering which allows to not take into account high frequency variations of the real wind. According to the regulator, the wind input is like a slow varying input with the colateral advantage to limit the power consumption. It can be considered that the simulated wind input and the real one (regulator point of view) are similars. The control commands (sail and rudder) are smooth according to the wind and sea; this point is also confirmed by the low power consumption logged during the experiment which was one of the control objectives. As for the wind input, it means that the sea state is also filtered by the mechanical inerta of the boat and the control laws. Autonomy is proven. Except for one team intervention for the submarine collision avoidance, the robot does not need any assistance and the power consumption remains low as the 12Ah has not been used. As the followed path is similar in the experiment and in the simulation, it is confirmed that the HWIL simulation provided a valuable status of feasibility for the experimentation. This point is fundamental

8 for mission preparation and allows to save time and money in experiments. Last point does not concern the comparison between but it is important to note that the embedded system is proven as robust. Indeed the mission has been performed in rough conditions of wind and sea and all the electronic devices have remained perfect. 5. CONCLUSIONS This paper has presented the control algorithms for an autonomous sailboat inspired from the physical behavior of a sailboat; it has been formalized with a static feedback for the sail and for the rudder; The static feedbacack approach is interesting as it limits the embedded calculation in comparison with other controllers. Moreover, as it is based on the human behaviour, it makes easier the interpretation of the controller. In the case of the use of these algorithms for an automatic pilot for yatching, as it is understandable, it remains acceptable. A work is on going to use this algorithm for disable people. The model validation is also a result of this experiment which leads to a mission preparation tool for sailboats. Each Vaimos mission for data acquisition is prepared with HWIL simulations and a complete log analysis. ACKNOWLEDGEMENTS The sailboat robot Vaimos is the result of a collaboration between LPO (Laboratoire de Physique des Oceans), RDT (Recherches et Developpements Technologiques) of IFRE- MER (Institut Francais de Recherche pour l Exploitation de la Mer) and ENSTA Bretagne (Ecole Nationale Superieure de Techniques Avancees). The author thanks LPOandRDTforhavingmadeavailabletheVaimosrobot for the test mission. REFERENCES M.L. Van Aatrijk, C.P. Tagliola, and P.W. Adriaans. Learning to sail. In Proceedings of 15th European Conference on Artificial Intelligence, 23. K.J. Astrom and R.M. Murray. Feedback Systems: An Introduction for Scientists and Engineers. Princeton University Press, 211. Morten Breivik and Thor I Fossen. Applying missile guidance concepts to motion control of marine craft. In Control Applications in Marine Systems, volume 7, pages , 27. Y. Briere. Iboat : an autonomous robot for long-term offshore operation. In Proceedings of 14th IEEE Mediterranean Electrotechnical Conference, 28. Collective. Proceedings of the 2nd International Robotic Sailing Conference. Universidade do Porto, 29. N. Cruz and J. Alves. Auto-heading controller for an autonomous sailboat. In Proceedings of MTS/IEEE conference OCEANS 1, 21. G. Elkaim. System identification based control of an unmanned autonomous wind-propelled catamaran. Control Engineering Practice, 17: , 29. T.I. Fossen. Guidance and control of Ocean vehicles. Wiley, Goulven Guillou. Architecture multi-agents pour le pilotage automatique des voiliers de compétition et extensions algébrique des réseaux de Petri. PhD thesis, Brest (UBO), November 21. P.Herrero,L.Jaulin,J.Vehi,andM.A.Sainz. Guaranteed set-point computation with application to the control of a sailboat. International Journal of Control Automation and Systems, 8, 21. N. Imbert and B. Clement. Launcher attitude control: some answers to the robustness issue. In Proceedings of 16th IFAC Symposium on Automatic Control in Aerospace, Saint-Petersburg, Russia, 24. L. Jaulin. Modelisation et commande dun bateau a voile. In Proceedings of 3rd Conference Internationale Francophone dautomatique, Douz,Tunisie, 24. L. Jaulin, B. Clement, Y. Gallou, F. Le Bars, O. Menage, O. Reynet, and J. Sliwka. Suivi de route pour un robot voilier. In Proceedings of 7th Conference Internationnale Francophone d Automatique, Grenoble, France, 212. O. Menage, F. Gaillard, T. Gorgues, T. Terre, P. Rousseaux, S. Prigent, Y. Auffret, L. Dussud, B. Forest, M. Repecaud, L. Jaulin, B. Clement, Y. Gallou, and F. Le Bars. VAIMOS: Voilier autonome instrument pour mesures ocanographiques de surface. In Symposium on Vulnerability of coastal ecosystems to global change and extreme events, Biarritz, France, 211. M. Neal. A hardware proof of concept of sailing robot for ocean observation. IEEE Journal of Oceanic Engineering, 31: , 26. M.A. Romero-Ramirez. Contribution a la commande de voiliers robotises. PhD thesis, Universite Pierre et Marie Curie, 212. P.F. Rynne and K.D. von Ellenrieder. Unmanned autonomous sailing: Current status and future role in sustained ocean observations. Marine Technology Society Journal, 43:21 3, 29. C. Sauze and M. Neal. An autonomous sailing robot for ocean observation. In Proceedings of TAROS 26, pages , Guildford, UK, 26. A. Schlaefer and O. Blaurock. Proceedings of the 4th International Robotic Sailing Conference. Springer, 211. R. Stelzer and K. Jafarmadar. History and recent developments in robotic sailing. In Proceedings of the 4th International Robotic Sailing Conference, Lubeck, Germany, 211. Roland Stelzer. Autonomous Sailboat Navigation. PhD thesis, De Montfort University, Leicester, 212. G. Thomas, O. Menage, T. Terre, and F. Gaillard. An innovative approach of the surface layer sampling. Journal des Sciences Halieutique et Aquatique, pages 15 19, 211. W.H. Warden. A control system model for autonomous sailboat navigation. In Proceedings of IEEESoutheastcon 91, volume 2, pages , E. Yeh and B.J.C. Binx. Fuzzy control for self-steering of a sailboat. In Proceedings of Singapore International Conference on Intelligent Control and Instrumentation, 1992.